Noise in the visual system may be early

1942, 1948) have been criticized on the grounds that the observer's overall efficiency, from stimulus to decision, is strongly dependent on the experimental conditions (Barlow, 1962). However, the observer's overall efficiency depends not only on the amount of noise in the system. It also depends on the efficiency of the observer's algorithm for reducing the multidimensional noisy data to a single number upon which a detection decision may he. made. Our visual Most models of visual performance incorporate random variations: noise. Sometimes the model maker injects this randomness as an afterthought, to account for trial-to-trial variations in subjects ' responses, Sometimes the noise is an essential part of the model. as in probability summation, where the key idea is the random detection that can occur in any of many independent channels (e.g.,Graham, 1977). Recent psychophysical evidence indicates that the main source of noise limiting visual detection is in the proximal stimulus: the photon noise arising from the random nature of light absorption (Banks,Geisler & Bennett, 1987; Cohn